Introduction: Non-Cartesian acquisitions are commonly used for MR imaging of short-T2 species, such as 23-sodium or 17-oxygen where an ultra-short echo time is essential for quantification, and for rapid 2-D and 3-D MR imaging of anatomy, physiology, or function. Since the Fast Fourier Transform (FFT) cannot be directly applied to non-Cartesian kspace data, alternate methods must be used for image reconstruction. Gridding-based image reconstruction allows the FFT to be used after interpolating the non-Cartesian data onto a Cartesian grid [1]. However, despite the fact that the algorithmic complexity of gridding is O(N) (Big-O notation), this step can still take significant time when there are millions of sample points or when gridding is performed multiple times as part of a multi-frequency reconstruction. We propose a method to perform gridding using a graphics processor (GPU) to achieve up to 29X acceleration for 3-D gridding.